一种卷积神经网络结合特征融合的网络入侵检测方法OA北大核心CSTPCD
A NETWORK INTRUSION DETEDTION METHOD BASED ON CONVOLUTIONAL NEURAL NETWORK AND FEATURE FUSION
为解决传统网络入侵检测方法中攻击特征过少、数据不平衡及模型收敛速度慢的问题,提出基于卷积神经网络结合特征融合的网络入侵检测方法.将流量数据转为灰度图像提取其纹理特征,再将纹理特征与流量特征进行特征融合以增加攻击特征量.使用Borderline-SMOTE方法对UNSW-NB15数据集进行数据平衡.运用逐层贪婪训练方法优化卷积神经网络模型提高模型的收敛速度.实验表明,该方法的性能优于其他检测方法,能将准确率最高提升到96.38%.
In order to solve the problems of few attack features,data imbalance and slow convergence in traditional network intrusion detection methods,this paper proposes an intrusion detection method based on convolutional neural network and feature fusion.This method converted the traffic data into a gray image to extract its texture features,and fused the texture features with network traffic features to increase the amount of attack characteristics.The Borderline-SMOTE method was used to balance the UNSW-NB15 data set.The greedy layer-wise training method was used to optimize the convolutional neural network model to improve the convergence speed of the model.Experiments show that the performance of this method is better than other detection methods,and the accuracy rate can be increased to 96.38%.
王雪妍;温蜜;李晋国;熊赟
上海电力大学计算机科学与技术学院 上海 200090复旦大学计算机科学技术学院 上海 200433
计算机与自动化
入侵检测特征融合逐层贪婪训练卷积神经网络Borderline-SMOTE
Intrusion detectionFeature fusionGreedy layer-wise trainingConvolutional neural networksBorderline-SMOTE
《计算机应用与软件》 2024 (008)
359-366 / 8
国家自然科学基金项目(61872230,61802248,61802249);上海市2019年度"科技创新行动计划"高新技术领域项目(19511103700);上海市科委项目(20020500600).
评论